Apptronik's Robot Park will feed Apollo 2 data into Google DeepMind's Gemini Robotics

Why it matters

Robot Park makes Apptronik's humanoid push a data play as much as a hardware rollout: Google gets real-world training loops, while Apptronik gets a route to smarter commercial robots.

Google DeepMind (@GoogleDeepMind) on July 6th put fresh weight behind Jeff Cardenas' Apptronik (@Apptronik), saying in a post on X that real-world data collected by Apptronik's latest Apollo 2 humanoid platform will help train and advance Gemini Robotics.

https://x.com/GoogleDeepMind/status/2074157282477154597

The post follows Apptronik's June 30th announcement that it had opened its newly expanded Robot Park, a nearly 90,000-square-foot data collection and training facility in Austin. Apptronik says Apollo 2 robots are already operating across Robot Park, customer sites and partner sites, collecting data from tasks in logistics, manufacturing, retail and other customer-driven settings.

The update clarifies the job Robot Park is meant to do. Apptronik is selling Apollo as a humanoid labor platform, but the facility is also a data factory for Google DeepMind's robotics models. Apptronik says Apollo 2 robots generate training data through teleoperation, autonomous execution and high-fidelity physics simulations. That dataset, according to Apptronik, is used to train and refine Gemini Robotics, Google DeepMind's family of AI models for physical-world tasks.

Cardenas, Apptronik's co-founder and CEO, has been building toward this from the University of Texas at Austin orbit rather than the AI lab circuit. Apptronik says it was founded in 2016 out of UT Austin's Human Centered Robotics Lab, after work on 15 robotic systems including NASA's Valkyrie. Cardenas' company bio says he studied technology commercialization at UT Austin, worked at the IC2 Institute and Deloitte Consulting, and previously founded several startups and a nonprofit. CTO and co-founder Dr. Nick Paine came from UT Austin's electrical and computer engineering program and worked on the NASA-JSC DARPA Robotics Challenge team that developed controllers for Valkyrie.

That history matters because Apptronik's pitch to Google is hardware experience, not just a humanoid shell. Google DeepMind introduced Gemini Robotics in March 2025 as a Gemini 2.0-based vision-language-action model that adds physical actions as an output modality for directly controlling robots. In the same announcement, Google DeepMind said it was partnering with Apptronik to build the next generation of humanoid robots with Gemini 2.0.

Google DeepMind also said in that March 2025 post that Gemini Robotics was trained primarily on data from the ALOHA 2 bi-arm platform and could be specialized for Apptronik's Apollo humanoid. In June 2025, Google DeepMind followed with Gemini Robotics On-Device, an on-device robotics model that it said could be adapted to Apptronik's Apollo humanoid after being trained only for ALOHA robots. The new Robot Park link gives Google DeepMind a way to get data from a more industrially relevant body: a humanoid platform operating around real facility layouts, customer workflows and human operators.

Apptronik's June 30th announcement says Apollo 2 comes in bipedal and wheeled-base configurations. The wheeled version is designed to fit within existing industrial mobile robot safety standards, while the bipedal version gives Apptronik a way to keep refining walking in more complex environments. Apollo 2's product page says the robot uses Apptronik's patented actuator technology, swappable batteries and fleet-management software called Fleet Connect.

Apptronik is already positioning Apollo 2 as a bridge to Apollo 3, which it describes as the next-generation commercial product. The company says Apollo 2 has been the workhorse behind Robot Park for more than a year and that its data streams will feed the development of Apollo 3. That puts Robot Park at the center of Apptronik's path from prototypes and pilots toward a fleet that can be sold into repetitive industrial work.

The financing behind that bet is substantial. In February 2025, Apptronik announced a $350 million Series A co-led by B Capital and Capital Factory, with participation from Google. The company said the round would support Apollo deployment, scaled operations and customer demand across automotive, electronics manufacturing, third-party logistics, beverage bottling, fulfillment and consumer packaged goods. The announcement named Mercedes-Benz and GXO Logistics as commercial agreements and said Apptronik had previously raised $28 million.

Google's role cuts both ways for Apptronik. Google gives Apptronik an AI partner with frontier-model resources and a robotics team that is trying to move Gemini beyond screens. Apptronik gives Google a fleet of human-scale machines that can collect the kind of messy physical data that internet-scale text and image corpora cannot provide. The unanswered commercial question is how much of the intelligence trained on Apptronik data stays uniquely useful to Apptronik if Gemini Robotics becomes a broader platform for other robotics companies.

That is the strategic tension under the July 6th post. Apptronik needs enough model advantage to make Apollo useful in customer sites before humanoid robotics becomes another hardware margin fight. Google DeepMind needs enough real-world embodiment data to prove Gemini can handle work outside labs and demo videos. Robot Park is where both sides are trying to turn repetitive industrial tasks into model-improving data.